How we grade every alert

When an alert fires we store it with a timestamp, the ticker, its direction (bullish or bearish) and, for options alerts, the exact contract. Nothing is added or edited afterwards, and alerts are never deleted because they lost.

At the same moment we record an opening price from a live quote: the midpoint of the bid and ask when there is one, otherwise the last trade. That price is the starting point for the grade.

Each alert is graded at the close of the trading session it belongs to. Alerts that fire after the 4 PM ET close, on a weekend or on a holiday roll into the next session and are graded at that session's close.

The closing price comes from a real quote at the close (the bid/ask midpoint, otherwise the last trade). For older history that we graded after the fact, we use the official daily closing price.

The grade is simple and the same for every alert:

• Bullish alerts win if the price closed higher than the opening price.
• Bearish alerts win if the price closed lower than the opening price.
• Anything else is a loss, including a close exactly at the opening price.

For an options alert we grade the option's own price, so time decay and changes in implied volatility count against it just as they would against a trader holding the contract.

A close-to-close grade misses what happened in between, so we also record each alert's path through the session: the peak favorable move (how far it went your way) and the worst adverse move (how far it went against you). For context we store the VIX at the time of the alert and SPY's return that day, so results can be compared with simply holding the market.

If we can't get a real price for an alert, at the open or at the close, we don't estimate one. The alert is marked as not gradable and left out of the win and loss counts rather than guessed. This mostly happens with thinly traded options contracts.

Grading is the feedback loop for the whole system. Each week we look at results by alert type, retire the ones that don't hold up and retune the rest. New models are trained on about 2 years of market history and only replace the current one if they do better on days they have never seen (walk-forward testing; see how walk-forward testing works).

We would rather you know these than find them out later:

• Midpoint prices are ideal fills. Real orders often fill worse than the midpoint, especially on options with wide spreads, and the grade includes no commissions or fees.
• One close is one horizon. Many setups take days or weeks to play out. A same-day grade can mark a trade as a loss that later worked, or a win that later reversed.
• A grade is not a trade. It doesn't include your stop, your size or your exit, and past results don't predict future ones.

That is why we publish every result instead of a single headline number: you can see the whole record and decide how much weight to give it.